API
< Back to all clusters
[TECHNOLOGY] · China · 2 sources

started · updated

XPENG researchers develop GroundingPI visual perception model

Researchers affiliated with XPENG, Peking University, and the University of Hong Kong have developed GroundingPI, an open-source foundational model designed to enhance how machines perceive the physical world. With approximately 4 billion parameters, the model focuses on visual perception, object localization, and spatial relationship comprehension, which are critical for robotics and autonomous vehicles.

In testing, GroundingPI outperformed 44 reference models across 34 public benchmarks. It achieved an average score of 73.68, surpassing the 71.54 score recorded by GPT-6 Astra in these specific visual perception tasks. The model distinguishes itself by moving beyond simple object recognition to precise spatial identification, using coordinates, points, and bounding boxes to locate specific parts of objects, such as a cup handle.

The project aims to bridge the gap between digital AI and Physical AI by providing a robust foundation for visual understanding that can later be used to learn robotic actions.

Entities

GPT-6 Astra · GroundingPI · Peking University · University of Hong Kong · XPeng